EP2642398B1 - Prélecture coordonnée dans des processeurs en mémoire cache hiérarchique - Google Patents

Prélecture coordonnée dans des processeurs en mémoire cache hiérarchique Download PDF

Info

Publication number
EP2642398B1
EP2642398B1 EP20130159754 EP13159754A EP2642398B1 EP 2642398 B1 EP2642398 B1 EP 2642398B1 EP 20130159754 EP20130159754 EP 20130159754 EP 13159754 A EP13159754 A EP 13159754A EP 2642398 B1 EP2642398 B1 EP 2642398B1
Authority
EP
European Patent Office
Prior art keywords
prefetch
cache
stream
unit
level
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
EP20130159754
Other languages
German (de)
English (en)
Other versions
EP2642398A1 (fr
Inventor
Brian P. Lilly
Hari S. Kannan
Gerard R. Williams III.
Mahnaz Sadoughi-Yarandi
Perumal R. Subramoniam
Pradeep Kanapathipillai
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Apple Inc
Original Assignee
Apple Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Apple Inc filed Critical Apple Inc
Publication of EP2642398A1 publication Critical patent/EP2642398A1/fr
Application granted granted Critical
Publication of EP2642398B1 publication Critical patent/EP2642398B1/fr
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0862—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches with prefetch
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F8/00—Arrangements for software engineering
    • G06F8/40—Transformation of program code
    • G06F8/41—Compilation
    • G06F8/44—Encoding
    • G06F8/443—Optimisation
    • G06F8/4441—Reducing the execution time required by the program code
    • G06F8/4442—Reducing the number of cache misses; Data prefetching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/30003—Arrangements for executing specific machine instructions
    • G06F9/3004—Arrangements for executing specific machine instructions to perform operations on memory
    • G06F9/30047—Prefetch instructions; cache control instructions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F12/00—Accessing, addressing or allocating within memory systems or architectures
    • G06F12/02—Addressing or allocation; Relocation
    • G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
    • G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
    • G06F12/0893—Caches characterised by their organisation or structure
    • G06F12/0897—Caches characterised by their organisation or structure with two or more cache hierarchy levels
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/10—Providing a specific technical effect
    • G06F2212/1016—Performance improvement
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/60—Details of cache memory
    • G06F2212/6024—History based prefetching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/60—Details of cache memory
    • G06F2212/6026—Prefetching based on access pattern detection, e.g. stride based prefetch
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
    • G06F2212/60—Details of cache memory
    • G06F2212/6028—Prefetching based on hints or prefetch instructions

Definitions

  • the present invention relates generally to cache performance, and in particular to methods and mechanisms for prefetching data in processors with multiple levels of caches.
  • Memory latency is frequently a large factor in determining the performance (e.g., instructions executed per second) of a processor in a given system.
  • the operating frequencies of processors have increased dramatically, while the latency for access to dynamic random access memory (DRAM) in the typical system has not decreased at the same rate. Accordingly, the number of processor clocks required to access the external memory has increased. Therefore, techniques for compensating for the relatively low speed of memory devices have been developed.
  • One technique is caching data in one or more caches located close to the processor. Caches are relatively small, low latency memories incorporated into the processor or coupled nearby.
  • processors typically use caches to combat the effects of memory latency on processor performance.
  • One way to mitigate the increasing latency of memory accesses is to prefetch data into a cache.
  • the term "prefetch" may generally refer to the fetching of data from memory before that data is actually needed for computation by instructions in the program.
  • One way that the memory bandwidth may be effectively utilized is to predict the information that will be accessed soon and then prefetch that information from the memory system into the cache. If the prediction is correct, the information may be a cache hit at the time of the actual request and thus the effective memory latency for actual requests may be decreased. On the other hand, if the prediction is incorrect, the prefetched information may replace useful information in the cache, causing more cache misses to be experienced than if prefetching were not employed and thus increasing the effective memory latency.
  • Certain types of computer programs process a long sequence of data where each element in the sequence is accessed only once. This type of access pattern usually results in cache misses since the required data is not in the cache at the time it is needed. This type of access may be referred to as a "data stream” or “stream”, which is prevalent in certain multimedia applications. Prefetching data based on a prediction of the stream may help prevent cache misses and improve processor efficiency.
  • the simplest type of prefetch prediction is a unit stride prediction.
  • a training mechanism may detect accesses to cache lines L and L+1. Therefore, the training mechanism may detect that the stride is 1, and so a prefetch unit may start prefetching cache lines L+2, L+3, etc.
  • other non-unit strides may be detected, and furthermore, strides may be to descending addresses instead of just ascending addresses.
  • Modern superscalar processors use very aggressive speculation techniques that involve reordering of memory accesses in order to achieve higher performance. Reordering of memory accesses results in obfuscation of any discernable pattern in the memory stream the further the operation progresses from the front-end of the machine.
  • lower-level caches have to contend with simultaneous request streams from multiple cores, which further increases the entropy of these access patterns. Some authors use the term lower-level cache to refer to caches closer to the core, while others use the term to refer to caches further from the core.
  • lower-level caches refers to caches further away from the core (e.g., L2 cache, L3 cache), while the term “upper-level cache” may refer to caches closer to the core (e.g., an L1 cache).
  • US 2009/199190 A1 discloses a processing system with a prefetch engine that uses a table identifying currently active streams.
  • US 6446167 B1 discloses a processing system including 3 cache levels.
  • a prefetch engine prefetches data into each of the cache levels here.
  • one centralized prefetch training mechanism per core may be utilized to train on streams.
  • the training information produced by the training mechanism may be utilized to generate prefetch requests at multiple levels of cache hierarchy.
  • Two or more prefetch units at separate cache levels may be coordinated utilizing the information obtained by the shared training mechanism.
  • the training mechanism may be located within a prefetch unit in the core, and the prefetch unit be coupled to the L1 cache.
  • the training mechanism may be configured to simultaneously train on a plurality of streams generated by the processor core. For a given stream, the training mechanism may monitor memory accesses of the stream and detect the stride of the stream. Then, the training mechanism may assign an identifier (ID) to the stream and indicate that the stream is live for prefetching.
  • a processor may include multiple cores, and each core may maintain a centralized training mechanism within its respective L1 prefetch unit. The lower level prefetch units may receive training information from each L1 prefetch unit.
  • the processor may include at least L1 and L2 caches and L1 and L2 prefetch units.
  • the L1 prefetch unit may send prefetch requests to the L2 cache, and the L1 prefetch unit may send a packet along with each prefetch request.
  • the packet may include a stream identifier (ID) and attributes associated with the prefetch request and the corresponding stream.
  • the attributes may include stride, page size, temporal/non-temporal status, load/store indicator, demand and prefetch addresses, and other information.
  • the L2 prefetch unit may utilize the attributes from the packet to generate prefetch requests.
  • the L2 prefetch unit may also utilize the attributes from the packet to determine how to process and store the retrieved prefetch data.
  • the L1 prefetch unit may maintain a table with information on the trained streams.
  • the L2 prefetch unit may also maintain a table with information on the trained streams.
  • the table maintained by the L2 prefetch unit may be a mirrored version of the table utilized by the L1 prefetch unit.
  • the L2 prefetch unit receives a prefetch request and accompanying packet from the core, the attributes from the packet may be stored in a corresponding entry in the table.
  • any additional lower level caches e.g., L3 may maintain tables synchronized to the tables of the upper level caches.
  • the L2 prefetch table may be the same size as the L1 prefetch table in a one-to-one mapping.
  • the L2 table may be as large as a combination of the L1 tables, and all of the entries of the two separate L1 tables may be stored in the L2 table.
  • the L2 table may be sized to store all of the entries from all of the L1 tables.
  • IC 10 includes a processor complex 20, memory controller 34, and memory physical interface circuits (PHYs) 36 and 38. It is noted that IC 10 may also include many other components not shown in FIG. 1 . In various embodiments, IC 10 may also be referred to as a system on chip (SoC), an application specific integrated circuit (ASIC), or an apparatus.
  • SoC system on chip
  • ASIC application specific integrated circuit
  • Processor complex 20 may include central processing units (CPUs) 22 and 26, level two (L2) cache 30, and bus interface unit (BIU) 32.
  • processor complex 20 may include other numbers of CPUs.
  • CPUs 22 and 26 may also be referred to as processors or cores.
  • CPUs 22 and 26 may include level one (L1) cache 24 and L1 cache 28, respectively.
  • L1 caches 24 and 28 may be coupled to L2 cache 30, and L2 cache 30 may be coupled to BIU 32, which may be coupled to memory controller 34.
  • Other embodiments may include additional levels of cache (e.g., level three (L3) cache).
  • processor complex 20 and CPUs 22 and 26 may include other components not shown in FIG. 1 .
  • the CPUs 22 and 26 may include circuitry to execute instructions defined in an instruction set architecture. Specifically, one or more programs comprising the instructions may be executed by the CPUs 22 and 26. Any instruction set architecture may be implemented in various embodiments. For example, in one embodiment, the PowerPC TM instruction set architecture may be implemented. Other exemplary instruction set architectures may include the ARM TM instruction set, the MIPS TM instruction set, the SPARC TM instruction set, the x86 instruction set (also referred to as IA-32), the IA-64 instruction set, etc.
  • L2 cache 30 may be configured to cache instructions and data for low latency access by CPUs 22 and 26.
  • the L2 cache 30 may comprise any capacity and configuration (e.g. direct mapped, set associative).
  • L2 cache 30 may be configured as a set-associative, writeback cache that is fully inclusive of L1 caches 24 and 28 within CPUs 22 and 26, respectively.
  • L2 cache 30 may include a variety of structures configured to support cache functionality and performance.
  • L2 cache 30 may include a miss buffer configured to store requests that miss the L2, a fill buffer configured to temporarily store data, a writeback buffer configured to temporarily store dirty evicted data and snoop copyback data, and/or a snoop buffer configured to store snoop requests.
  • L2 cache 30 may include a crossbar (not shown) for managing data flow between L2 cache 30 and BIU 32 and between L2 cache 30 and CPUs 22 and 26.
  • L2 cache 30 may implement arbitration logic to prioritize cache access among various cache read and write requestors. Other variations of L2 cache 30 configurations are possible and contemplated.
  • L2 cache 30 may be coupled to memory controller 34 via BIU 32.
  • BIU 32 may also include various other logic structures to couple CPUs 22 and 26 and L2 cache 30 to various other devices and blocks.
  • Memory controller 34 may include any number of memory ports and may include circuitry configured to interface to memory.
  • the memory controller 34 may be configured to interface to dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM), double data rate (DDR) SDRAM, DDR2 SDRAM, Rambus DRAM (RDRAM), etc.
  • DRAM dynamic random access memory
  • SDRAM synchronous DRAM
  • DDR double data rate SDRAM
  • DDR2 SDRAM Rambus DRAM
  • RDRAM Rambus DRAM
  • Memory controller 34 may also be coupled to memory physical interface circuits (PHYs) 36 and 38.
  • Memory PHYs 36 and 38 are representative of any number of memory PHYs which may be coupled to memory controller 34.
  • the memory PHYs 36 and 38 may be configured to interface to memory devices (not shown).
  • embodiments may include other combinations of components, including subsets or supersets of the components shown in FIG. 1 and/or other components. While one instance of a given component may be shown in FIG. 1 , other embodiments may include two or more instances of the given component. Similarly, throughout this detailed description, two or more instances of a given component may be included even if only one is shown, and/or embodiments that include only one instance may be used even if multiple instances are shown.
  • Core 40 is one example of a processor core, and core 40 may be utilized within a processor complex, such as processor complex 20 of FIG. 1 .
  • each of CPUs 22 and 26 of FIG. 1 may include the components and functionality of core 40.
  • Core 40 may include fetch and decode (FED) unit 42, map and dispatch unit 46, execution units 48, load/store unit (LSU) 50, memory management unit (MMU) 60, and core interface unit (CIF) 62. It is noted that core 40 may include other components not shown in FIG. 2 .
  • FED fetch and decode
  • LSU load/store unit
  • MMU memory management unit
  • CIF core interface unit
  • FED unit 42 which includes the L1 instruction cache 44, may be configured to fetch instructions from memory (or the L2 cache) and decode the fetched instructions. Decoded instructions may be conveyed to the map and dispatch unit 46. Map and dispatch unit 46 may be configured to map decoded instructions to physical registers. Map and dispatch unit 46 may also be configured to dispatch instructions to execution units 48 and LSU 50. Execution units 48 may include any number and type of execution units (e.g., integer, floating point, vector).
  • CIF 62 may be coupled to L1 data cache 52, FED unit 42, and MMU 60.
  • CIF 62 may be configured to manage the interface between core 40 and the L2 cache.
  • MMU 60 may be configured to perform address translation and memory management functions.
  • LSU 50 may include L1 data cache 52 coupled to prefetch unit 54.
  • Prefetch unit 54 may include prefetch training mechanism 56 and prefetch training table 58.
  • LSU 50 may also be coupled to an L2 cache (not shown) via CIF 62. It is noted that LSU 50 may also include other components (e.g., load queue, store queue) not shown in FIG. 2 .
  • Prefetch unit 54 may be configured to generate prefetch requests in order to prefetch data into the L1 data cache 52.
  • prefetch unit 54 may be configured to concurrently maintain multiple separate, independent prefetch streams.
  • a "prefetch stream” may refer to a stream of addresses and blocks associated with those addresses that are prefetched into the cache as a result of a detected prefetch pattern.
  • the prefetch pattern may describe the difference between consecutive memory accesses in the prefetch stream. This difference between consecutive memory accesses may be referred to as the "stride" of the prefetch stream.
  • Training mechanism 56 may monitor the load/store traffic through L1 data cache 52 and/or at other points in the core 40. Training mechanism 56 may detect patterns of access in the load/store traffic and identify streams for prefetching. For each separate stream, training mechanism 56 may detect unit stride patterns for prefetches, such as prefetches of consecutive blocks in memory, either in an increasing-address or decreasing-address direction. Other embodiments may detect more complex prefetch patterns, including larger strides and repeating patterns that are not a single fixed distance between memory accesses.
  • instructions being executed by core 40 may proceed in order until the instructions get to the execution units 48, at which point the instructions may be reordered. At this point, garbling may occur making it difficult to train on an instruction stream.
  • Training mechanism 56 may train on instructions in core 40 prior to the instructions being reordered. Then, the training information may be propagated to the lower levels of the cache hierarchy.
  • Prefetch training mechanism 56 may train on streams generated by core 40. Training may include detecting and analyzing the memory accesses associated with the various streams to determine information associated with the memory accesses. When prefetch training mechanism 56 has trained on a stream, it may assign a stream identifier (ID) to the stream and store an entry in table 58 with the stream ID. The entry may include attributes associated with the stream, such as temporality or non-temporality of data, stride, page size, memory access type, demand address, prefetch address, as well as other information.
  • ID stream identifier
  • the entry may include attributes associated with the stream, such as temporality or non-temporality of data, stride, page size, memory access type, demand address, prefetch address, as well as other information.
  • Prefetch unit 54 may include a table 58 having a plurality of entries for storing training information on a plurality of streams.
  • a given prefetch stream may be assigned to one of the entries of table 58, and generation of prefetches may be performed for that prefetch stream based on the contents of the entry. For example, the addresses of the prefetch request may be compared to the current address and/or the next expected address in the table entry.
  • prefetch unit 54 may maintain a least recently used (LRU) status across the entries in table 58 for replacement purposes.
  • LRU least recently used
  • a stream may be associated with a unique stream identifier (ID), and each entry of table 58 may include a stream ID along with attributes associated with a given stream.
  • the stream ID and attributes may be propagated from the core down to the L2 and lower level caches.
  • core 40 When core 40 sends out a prefetch request to the L2 cache, it may send data including the stream ID and other attributes with (or in association with) the request.
  • the conveyed ID and attributes may take the form of a tuple, or otherwise, and may be included in an associated packet(s).
  • packet is used for ease of discussion and is not intended to be limited to a particular size or configuration of data. Rather, the methods and mechanisms described herein may convey the "packet" related data in many of a variety of ways. Conveying training data and other attributes may take any form or format.
  • the tuple may contain the prefetch address, the stream ID, stride (e.g., 2 lines), and other attributes.
  • the L2 prefetch unit may store the attributes of the tuple in the corresponding entry in a L2 prefetch table (not shown). Then, the L2 prefetch unit may compare the new tuple to the last request the core sent to make sure that it stays ahead of the addresses being fetched by the L1 prefetch unit. If the stream ID in the tuple is a new stream ID, then the L2 prefetch unit may create a new entry for the specific stream ID.
  • the prefetching may extend forward to prefetch further into the predicted memory locations of the stream.
  • the L1 prefetch unit may add a first start distance to the first prefetch request, and the L2 prefetch unit may add a second start distance on top of the first start distance.
  • Lower level prefetch units e.g., L3 prefetch unit
  • the start distance may be programmed locally at each prefetch unit.
  • the start distance may be sent as one of the attributes within the tuple.
  • the L2 prefetch unit may not do any training itself.
  • the L2 prefetch unit may get setup by the L1 prefetch unit, and then the L2 prefetch unit may generate more prefetch requests further along in the stream than the L1 prefetch unit.
  • the L2 prefetch unit may be an extension of the L1 prefetch unit. This approach may be extended to a third level, fourth level, and so on, where regardless of the number of hierarchical levels of caches, only a single unified training mechanism may be utilized. In such an embodiment, the L3 prefetch unit would be a slave to the upper two prefetch units.
  • L1 and L2 prefetch units are shown coupled to L2 cache 78, and core 72 may include a L1 prefetch unit 74. It is noted that core 72 may also include many other components that are not shown in FIG. 3 for purposes of clarity.
  • L1 prefetch unit 74 may include a prefetch table 76 for storing data associated with streams that have been trained.
  • L2 cache 78 may include L2 prefetch unit 80, which may include prefetch training table 82.
  • the prefetch requests and accompanying training data may pass through one or more components (e.g., MMU, CIF) between L1 prefetch unit 74 and L2 cache 88.
  • core 72 may issue a prefetch request to L2 cache 78.
  • the prefetch request may include a memory address of a line to be accessed.
  • training data may be sent along with the prefetch request.
  • the training data may identify the stream ID of the prefetch request, and may include other attributes (stored in table 76) associated with that particular stream ID.
  • L2 prefetch unit 80 may compare the training data to the entry in the table corresponding to the specific stream ID. For the example shown in FIG. 3 , the stream ID may be '2', and L2 prefetch unit 80 may use this stream ID to locate the corresponding entry in table 82. L2 prefetch unit 80 may retrieve one or more attributes (e.g., stride) which may be used by L2 prefetch unit 80 to generate prefetch requests to memory (or to a lower level cache). In effect, the training data sent along with the prefetch requests may allow the entries of tables 78 and 82 to be synchronized.
  • attributes e.g., stride
  • tables 76 and 82 may not contain exactly the same information, but tables 76 and 82 may generally be synchronized.
  • table 76 may be updated for a particular stream ID and there may be a time lag before table 82 is updated for that particular stream ID.
  • one or more of the fields of the entries in table 76 and 82 may contain different information and/or may be formatted differently.
  • an address field corresponding to the last generated prefetch request may contain a different address in the entries of table 76 as compared to the entries of table 82.
  • the data contained in table 82 may be reflective of the data in table 76 and may in this sense be synchronized to the data in table 76.
  • each packet sent from core 72 to L2 cache 78 may include a bit which indicates if the stream is a new stream. If the stream is a new stream, as indicated by this bit, then the L2 prefetch unit 80 may find the corresponding entry in table 82 and clear this entry. Then, L2 prefetch unit 80 may fill the attributes of the cleared entry with the information contained in the packet. If the stream is an existing stream, then the packet may contain at least the stream ID, and L2 prefetch unit 80 may retrieve the corresponding entry corresponding to this stream ID in table 82.
  • L2 prefetch unit 80 may generate new prefetch requests based on the attributes stored in the corresponding entry in table 82. The new prefetch requests issued by L2 prefetch unit 80 may allow it to stay ahead of core 72 for this particular stream so that future requests will result in hits in the L2 cache 78.
  • L2 prefetch unit 80 may assume that each received packet and prefetch request is for a new stream. Therefore, in this embodiment, table 82 may be updated on any prefetch request that is received by L2 cache 78. For example, L2 prefetch unit 80 may overwrite the data in the corresponding entry in table 82 with the information in the received packet. In various embodiments, the L2 prefetch unit 80 may assume that any received prefetch request is for a new stream unless the request includes the same stream information as an existing stream.
  • the packet may include information that controls the placement of data and the replacement policies in L2 cache 78. If a particular stream has a temporal or non-temporal attribute, L1 prefetch unit 74 may convey this information to L2 prefetch unit 80. In one embodiment, L2 prefetch unit 80 and the lower level caches may use this information to determine which way of the cache to store the corresponding data for this particular stream.
  • L2 cache 78 may utilize the addresses and information from a prefetch request and packet sent from core 72 to make sure L2 prefetch unit 80 stays a predetermined number of lines ahead of the address contained in the prefetch request. If L2 prefetch unit 80 determines it has lagged behind, then L2 prefetch unit 80 may jump forward to stay ahead of L1 prefetch unit 74. L2 prefetch unit 80 may be configured to stay ahead of L1 prefetch unit 74 in the stream by a certain distance.
  • Prefetch training table 90 may be utilized to store information associated with various trained streams.
  • Table 90 is representative of a table which may be utilized by a prefetch unit at any level of the overall cache hierarchy.
  • the number of entries in table 90 may vary depending on the embodiment, the level of the cache hierarchy at which table 90 is located, the maximum number of concurrent streams, as well as other information.
  • training table 90 may include other information, and/or some of the attributes shown in table 90 may not be included in the table. Alternatively, one or more of the attributes shown in table 90 may be stored in another location of the corresponding prefetch unit.
  • Each entry of table 90 may include a plurality of attributes.
  • a stream ID 92 may be included in each entry and may be utilized to identify the entry.
  • the stream ID 92 may correspond to a respective stream ID that has been trained by a training mechanism in the core.
  • the stride 94 may be the detected stride for the stream, and in one embodiment may be a number of lines.
  • the temporality indicator 96 may be a bit that indicates if the stream corresponds to temporal or non-temporal data.
  • the L2 cache and lower level caches may utilize temporality indicator 96 to bias the way selection in its LRU mechanism. The value of temporality indicator 96 may also affect replacement policy within a cache.
  • Another attribute stored in the entries of table 90 may be the page size attribute 98 (e.g., 64 KB, 128 KB, 1 MB).
  • a L2 prefetch unit (and lower level prefetch units) may use this attribute to determine the page boundaries. When encountering a page boundary, the L2 prefetch unit may stop prefetching. The L2 prefetch unit may also use a prefetch count to determine when to stop prefetching, such that once the prefetch count has been reached, then the L2 prefetch unit may stop prefetching until it receives another prefetch request.
  • the prefetch count may be programmed or set separately from the attributes received with the packets accompanying prefetch requests. In one embodiment, the prefetch count may not be stored in table 90, although in other embodiments, the prefetch count may be stored in table 90.
  • a "turbo" mode may be utilized by the various prefetch units.
  • the prefetch count may be increased in order to stay further ahead of the given stream.
  • the status of the given stream may be tracked by a separate counter, and this separate counter may count the number of prefetch requests arriving at each cache level from above. For example, the number of L1 prefetch requests that have been received at the L2 level for a particular stream may be counted at the L2 prefetch unit.
  • the L2 prefetch unit may enter a turbo mode, which will increase the prefetch count and increase the number of prefetch requests generated by the L2 prefetch unit in an attempt to stay further ahead of the stream.
  • the threshold may be programmable and may vary from embodiment to embodiment.
  • the L1 prefetch unit may implement an analogous technique to apply turbo mode based on tracking the number of demand requests received at the L1 cache level for the various streams.
  • the value of the separate counter may or may not be stored in table 90, depending on the embodiment. Various embodiments including combinations of the above are possible and are contemplated.
  • the prefetch unit may not be sure if the next page in the physical address space corresponds to the next page in the program's virtual address space. Therefore, in one embodiment, when a lower level prefetch unit, such as an L2 prefetch unit, reaches a page boundary, the L2 prefetch unit may stop at the page boundary. When the higher level prefetch unit, in this case the L1 prefetch unit, crosses the page boundary, it may send a new prefetch request with the address of the next page to the L2 prefetch unit.
  • a lower level prefetch unit such as an L2 prefetch unit
  • Each entry of table 90 may also include a stream type attribute 100 to indicate if the stream is a load stream or a store stream. For example, if the stream is a load stream, then the L2 prefetch unit may retrieve lines of data in a shared state. If the stream is a store stream, then the L2 prefetch unit may bring the data into the core in an exclusive state so that the core can modify the data.
  • a L1 prefetch unit may combine information between load and store streams into a single entry in L1 prefetch training table. For example, if a load stream and a store stream are both targeting the same region of memory, the L1 prefetch unit may collapse the load and store streams into one stream. Therefore, the L1 prefetch unit may create one stream entry in the L1 table instead of creating two different stream entries. For example, in one embodiment, if the load stream is generated first, the L1 prefetch unit will train on the load stream and fill a corresponding entry in the L1 table with the attributes of the load stream.
  • the L1 prefetch unit determines that the store request is to the same address as a pending load request, then the store request may be collapsed into the load stream, and then the load stream may be converted into a store stream. Then, when a corresponding prefetch request gets sent out from the L1 to the L2, the accompanying packet may get sent out with a "store" attribute, which means that the L2 will be responsible for retrieving the data in an exclusive state, instead of in a shared state.
  • Each entry in table 90 may also include a demand address 102 attribute and a prefetch address 104 attribute.
  • the L1 cache When the L1 cache receives a demand request, it may store the address of the demand request in the demand address 102 attribute of the corresponding stream entry.
  • the L2 prefetch unit When an L2 (or lower level) prefetch unit receives a prefetch request from a higher level cache, the L2 prefetch unit may store the address of the prefetch request in the demand address 102 attribute.
  • the L1 prefetch unit and lower level prefetch units may store the current address being prefetched in the prefetch address 104 attribute of the corresponding entry.
  • the L2 prefetch unit may compare the prefetch request address with the demand address 102 stored in table 90 for that particular stream ID. If the incoming prefetch request is ahead of the physical address stored in that entry, then the L2 prefetch unit may overwrite the demand address 102 attribute of the entry with the new address.
  • the L2 prefetch unit may check to see if the prefetch request corresponds to an existing entry in the L2 table, and if so, it may advance that prefetch request. Otherwise, the L2 prefetch unit may flush an old entry from the table, and then the L2 prefetch unit may populate the entry with the new data contained in the prefetch request and the accompanying packet.
  • an IC or SoC may include two cores, core 106 and core 108. In other embodiments, other numbers of cores (e.g., four, eight) may be included in the IC or SoC.
  • Tables 112 and 114 are each shown as including four entries. In other embodiments, tables 112 and 114 may include other numbers of entries (e.g., eight, sixteen). Any number and type of attributes may be stored in tables 112, 114, and 116. For example, one or more of the attributes shown in table 90 (of FIG. 4 ) may be stored in tables 112, 114, and 116. Additionally, other attributes now shown in table 90 may be stored in tables 112, 114, and 116.
  • the first four entries of table 116 of L2 cache 112 may be synchronized with the entries of table 112.
  • the entries for the bottom four entries of table 116 of L2 cache 110 may be synchronized with the entries of table 114.
  • the entries in table 114 may be assigned stream IDs 1-4, and these stream IDs may be mapped to stream IDs 5-8 in table 116.
  • entries in table 116 may be synchronized with entries in tables 112 and table 114 as prefetch requests and accompanying packets are sent from the cores to L2 cache 110.
  • Table 116 may include enough entries to store and mirror all of the entries of all of the L1 caches that are coupled to L2 cache 110.
  • table 116 includes eight entries.
  • table 116 may include other numbers of entries. For example, in a processor with eight cores and with each core storing 16 entries for 16 separate stream IDs in a core prefetch training table, then table 116 may include 128 entries for storing all of the entries from all of the cores. Other numbers of cores coupled to an L2 cache are possible and are contemplated.
  • FIG. 6 a block diagram of one embodiment of L2 and L3 prefetch units is shown.
  • FIG. 6 illustrates a continuation of the block diagram illustrated in FIG. 3 .
  • L2 cache 78 may receive prefetch requests and training information from a core (e.g., core 72 of FIG. 3 ), and then L2 prefetch unit 80 may store training data in table 82.
  • L2 prefetch unit 80 may generate prefetch requests based on the received prefetch requests, training data, and/or the data in table 82.
  • L2 prefetch unit 80 may convey the prefetch requests to L3 cache 120, and L2 prefetch unit 80 may also generate and convey the training data along with the prefetch request.
  • the training data may include attributes from table 82 for the particular stream ID associated with the prefetch request.
  • L3 prefetch unit 122 may receive the prefetch request and training data and store the training data in the corresponding entry of table 124.
  • L3 prefetch unit 122 may function in a similar manner to L2 prefetch unit 80 in response to receiving a prefetch request and accompanying training data.
  • L3 prefetch unit 122 may utilize the training data and/or corresponding table 124 entry to stay ahead of the prefetch requests generated by L2 prefetch unit 80.
  • training information may be sent from the core to L2 cache 78 and to L3 cache 120 in parallel. In this way, L2 table 82 and L3 table 124 may be updated at the same time by the training information.
  • L3 prefetch table 124 may include as many entries as the combination of both L2 prefetch tables.
  • FIG. 7 one embodiment of a method for generating prefetch requests at multiple cache levels is shown. For purposes of discussion, the steps in this embodiment are shown in sequential order. It should be noted that in various embodiments of the method described below, one or more of the elements described may be performed concurrently, in a different order than shown, or may be omitted entirely. Other additional elements may also be performed as desired.
  • a training mechanism coupled to a first cache level may train on a given stream (block 130).
  • the training mechanism may be located within a prefetch unit within a processor core.
  • a processor may include multiple cores, and each core may have a single training mechanism.
  • the training information generated by the training mechanism may be stored in a corresponding entry of a first table at the first cache level (block 132).
  • the entry in the first table may include a stream ID for the trained stream.
  • the first cache level may be a L1 cache, and the first table may be stored in a prefetch unit coupled to the L1 cache.
  • training information may be conveyed from the first cache level to a second cache level (block 134).
  • a prefetch request may be generated by the prefetch unit at the first cache level and sent to the second cache level, and the training information may be included within a packet that accompanies the prefetch request.
  • a prefetch unit at the second cache level may receive the training information (block 136), and then the prefetch unit may issue prefetch requests based on the received training information (block 138).
  • the prefetch unit at the second cache level may issue prefetch requests that are a certain distance ahead of the prefetch requests being issued at the first cache level. After an initial lag, prefetch requests from the first cache level should begin to hit in the second cache level for the given stream.
  • the prefetch unit at the second cache level may also store the received training information in a corresponding entry in a second table (block 140). It is noted that blocks 138 and 140 may occur simultaneously or block 140 may be performed prior to block 138, depending on the embodiment.
  • the second cache level may be a L2 cache.
  • the prefetch unit at the second cache level may convey the training information to a third cache level (block 142).
  • the training information may be included within a packet that accompanies the prefetch request to the third cache level.
  • a prefetch unit at the third cache level may receive the training information (block 144), and then the prefetch unit at the third cache level may issue prefetch requests based on the received training information (block 146).
  • the received training information may be also stored in a corresponding entry in a third table (block 148).
  • the third cache level may be a L3 cache.
  • multiple streams may be trained in parallel, and multiple streams may be at various points within the block diagram of FIG. 7 at any given time.
  • Other embodiments may include other numbers of caches.
  • the training mechanism at a highest level of cache may share prefetch training information with the lower level caches.
  • system 150 may represent chip, circuitry, components, etc., of a desktop computer 160, laptop computer 170, tablet computer 180, cell phone 190, or otherwise.
  • the system 150 includes at least one instance of IC 10 (of FIG. 1 ) coupled to an external memory 152.
  • IC 10 is coupled to one or more peripherals 154 and the external memory 152.
  • a power supply 156 is also provided which supplies the supply voltages to IC 10 as well as one or more supply voltages to the memory 152 and/or the peripherals 154.
  • power supply 156 may represent a battery (e.g., a rechargeable battery in a smart phone, laptop or tablet computer).
  • more than one instance of IC 10 may be included (and more than one external memory 152 may be included as well).
  • the memory 152 may be any type of memory, such as dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate (DDR, DDR2, DDR3, etc.) SDRAM (including mobile versions of the SDRAMs such as mDDR3, etc., and/or low power versions of the SDRAMs such as LPDDR2, etc.), RAMBUS DRAM (RDRAM), static RAM (SRAM), etc.
  • DRAM dynamic random access memory
  • SDRAM synchronous DRAM
  • SDRAM double data rate SDRAM
  • RDRAM RAMBUS DRAM
  • SRAM static RAM
  • One or more memory devices may be coupled onto a circuit board to form memory modules such as single inline memory modules (SIMMs), dual inline memory modules (DIMMs), etc.
  • SIMMs single inline memory modules
  • DIMMs dual inline memory modules
  • the devices may be mounted with IC 10 in a chip-on-chip configuration, a package-on-package configuration, or
  • the peripherals 154 may include any desired circuitry, depending on the type of system 150.
  • peripherals 154 may include devices for various types of wireless communication, such as wifi, Bluetooth, cellular, global positioning system, etc.
  • the peripherals 154 may also include additional storage, including RAM storage, solid state storage, or disk storage.
  • the peripherals 154 may include user interface devices such as a display screen, including touch display screens or multitouch display screens, keyboard or other input devices, microphones, speakers, etc.
  • computer readable medium 200 may include any non-transitory storage media such as magnetic or optical media, e.g., disk, CD-ROM, or DVD-ROM, volatile or non-volatile memory media such as RAM (e.g. SDRAM, RDRAM, SRAM, etc.), ROM, etc., as well as media accessible via transmission media or signals such as electrical, electromagnetic, or digital signals, conveyed via a communication medium such as a network and/or a wireless link.
  • non-transitory storage media such as magnetic or optical media, e.g., disk, CD-ROM, or DVD-ROM, volatile or non-volatile memory media such as RAM (e.g. SDRAM, RDRAM, SRAM, etc.), ROM, etc., as well as media accessible via transmission media or signals such as electrical, electromagnetic, or digital signals, conveyed via a communication medium such as a network and/or a wireless link.
  • RAM e.g. SDRAM, RDRAM, SRAM, etc.
  • ROM etc.
  • the data structure(s) of the circuitry on the computer readable medium 200 may be read by a program and used, directly or indirectly, to fabricate the hardware comprising the circuitry.
  • the data structure(s) may include one or more behavioral-level descriptions or register-transfer level (RTL) descriptions of the hardware functionality in a high level design language (HDL) such as Verilog or VHDL.
  • HDL high level design language
  • the description(s) may be read by a synthesis tool which may synthesize the description to produce one or more netlists comprising lists of gates from a synthesis library.
  • the netlist(s) comprise a set of gates which also represent the functionality of the hardware comprising the circuitry.
  • the netlist(s) may then be placed and routed to produce one or more data sets describing geometric shapes to be applied to masks.
  • the masks may then be used in various semiconductor fabrication steps to produce a semiconductor circuit or circuits corresponding to the circuitry.
  • the data structure(s) on computer readable medium 200 may be the netlist(s) (with or without the synthesis library) or the data set(s), as desired.
  • the data structures may comprise the output of a schematic program, or netlist(s) or data set(s) derived therefrom.
  • computer readable medium 200 includes a representation of IC 10
  • other embodiments may include a representation of any portion or combination of portions of IC 10 (e.g., processor complex 20, CPU 22, L2 cache 30).

Claims (18)

  1. Un processeur (20) comprenant :
    une hiérarchie de caches comprenant au moins deux niveaux de cache ;
    une première unité de pré-obtention (74) associée à un premier cache à un premier niveau de cache de ladite hiérarchie de caches ;
    une seconde unité de pré-obtention (80) associée à un second cache à un second niveau de cache de ladite hiérarchie de caches ; et
    un mécanisme d'apprentissage de pré-obtention (56) associé au premier cache ;
    caractérisé en ce que le mécanisme d'apprentissage de pré-obtention est configuré pour générer des informations d'apprentissage pour chaque flux d'une pluralité de flux (130), la première unité de pré-obtention (74) étant configurée pour délivrer des requêtes de pré-obtention pour pré-obtenir des données correspondant à la pluralité de flux dans le premier cache, sur la base des informations d'apprentissage ;
    dans lequel la première unité de pré-obtention (74) est configurée pour transférer lesdites informations d'apprentissage à la seconde unité de pré-obtention (80) ; et
    dans lequel la seconde unité de pré-obtention (80) est configurée pour délivrer des requêtes de pré-obtention pour pré-obtenir des données correspondant à la pluralité de flux dans le second cache (138), sur la base des informations d'apprentissage.
  2. Le processeur de la revendication 1, dans lequel la seconde unité de pré-obtention (80) n'exécute aucun apprentissage de pré-obtention.
  3. Le processeur de la revendication 1, dans lequel lesdites informations d'apprentissage comprennent des données temporelles (96) qui sont utilisées par le second cache en tant que partie d'une politique de remplacement de cache.
  4. Le processeur de la revendication 3, dans lequel la première unité de pré-obtention est configurée pour :
    conserver une première table de pré-obtention (76) comprenant une pluralité de rubriques, chaque rubrique comprenant un identifiant de flux (ID) et des attributs associés à un flux correspondant ;
    générer une requête de pré-obtention pour un flux donné sur la base d'attributs stockés dans une rubrique correspondante de la première table de pré-obtention ; et
    transférer la requête de pré-obtention et un paquet vers le second cache, le paquet comprenant un ou plusieurs attributs provenant de la rubrique correspondante de la première table de pré-obtention.
  5. Le processeur de la revendication 4, dans lequel la seconde unité de pré-obtention est en outre configurée pour :
    détecter une frontière de page pour un flux donné ; et
    arrêter la pré-obtention en réponse à la détection de ladite frontière de page.
  6. Le processeur de la revendication 4, dans lequel la seconde unité de pré-obtention est configurée pour conserver une seconde table de pré-obtention (82) avec une rubrique pour chaque rubrique de la première table de pré-obtention, et dans lequel les informations d'apprentissage de la seconde table de pré-obtention sont synchronisées avec les informations d'apprentissage de la première table de pré-obtention.
  7. Le processeur de la revendication 1, dans lequel la seconde unité de pré-obtention est configurée pour :
    compter un nombre de requêtes de pré-obtention pour un flux particulier délivré par la première unité de pré-obtention et reçu par le second cache ; et
    augmenter un nombre de requêtes de pré-obtention générées par la seconde unité de pré-obtention en réponse à la détermination que ledit nombre dépasse un seuil.
  8. Un processeur selon la revendication 1,
    dans lequel le mécanisme d'apprentissage de pré-obtention est couplé à un niveau le plus élevé de cache de la hiérarchie de caches et génère les informations d'apprentissage pour une pluralité de flux, les informations d'apprentissage étant utilisées pour générer des requêtes de pré-obtention aux au moins deux niveaux de cache de ladite hiérarchie de caches.
  9. Le processeur de la revendication 8, dans lequel la pluralité de caches comprend au moins un cache de niveau un (L1) et un cache de niveau deux (L2).
  10. Le processeur de la revendication 8, où ledit premier niveau de cache est le plus haut niveau de cache de ladite hiérarchie de caches et la première unité de pré-obtention est couplée au niveau de cache le plus élevé, dans lequel la première unité de pré-obtention est configurée pour :
    générer une pluralité de requêtes de pré-obtention sur la base des informations d'apprentissage ;
    générer un paquet pour chaque requête de pré-obtention, le paquet comprenant des attributs associés à la requête de pré-obtention et à un flux correspondant ; et
    transmettre la requête de pré-obtention et le paquet au second cache au second niveau de cache, le second niveau de cache étant un niveau de cache inférieur de ladite hiérarchie de caches.
  11. Le processeur de la revendication 10, dans lequel chaque flux ayant fait l'objet d'un apprentissage possède un identifiant de flux associé (92), et dans lequel les attributs comprennent au moins l'identifiant de flux.
  12. Le processeur de la revendication 11, dans lequel les attributs comprennent en outre un pas d'adressage (94) du flux correspondant.
  13. Le processeur de la revendication 11, dans lequel l'unité de pré-obtention est en outre configurée pour attribuer un flux de charge et un flux de stockage à un identifiant de flux unique, en réponse à la détection d'accès à une même adresse par le flux de charge et le flux de stockage.
  14. Un procédé, comprenant :
    la génération d'informations d'apprentissage sur une pluralité de flux,
    l'apprentissage étant exécuté par un mécanisme d'apprentissage de pré-obtention (56) couplé à un premier cache à un premier niveau de cache d'une hiérarchie de caches comprenant au moins deux niveaux de cache, une première unité de pré-obtention (74) étant configurée pour délivrer des requêtes de pré-obtention pour pré-obtenir des données correspondant à la pluralité de flux dans le premier cache, sur la base des informations d'apprentissage ; et
    le transfert desdites informations d'apprentissage vers un second niveau de cache (78) de la hiérarchie de caches ;
    dans lequel une seconde unité de pré-obtention (80) couplée à un second cache audit second niveau de cache est configurée pour délivrer des requêtes de pré-obtention pour pré-obtenir des données correspondant à la pluralité de flux dans le second cache, sur la base des informations d'apprentissage.
  15. Le procédé de la revendication 14, dans lequel ladite seconde unité de pré-obtention n'exécute pas d'apprentissage de pré-obtention.
  16. Le procédé de la revendication 15, comprenant en outre :
    le stockage des informations d'apprentissage dans une première table au premier niveau de cache ; et
    le stockage des informations d'apprentissage dans une seconde table au second niveau de cache, la seconde table étant synchronisée avec la première table.
  17. Le procédé de la revendication 14, comprenant en outre :
    le comptage par la première unité de pré-obtention d'un nombre de requêtes de pré-obtention pour un flux particulier délivré par la première unité de pré-obtention et reçu par le second cache ; et
    l'augmentation d'un nombre de requêtes de pré-obtention générées par la seconde unité de pré-obtention en réponse à la détermination que ledit nombre dépasse un seuil.
  18. Le procédé de la revendication 17, dans lequel ladite information d'apprentissage comprend des données temporelles (96) qui sont utilisées par le second cache en tant que partie d'une politique de remplacement de cache.
EP20130159754 2012-03-20 2013-03-18 Prélecture coordonnée dans des processeurs en mémoire cache hiérarchique Active EP2642398B1 (fr)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US13/425,123 US9098418B2 (en) 2012-03-20 2012-03-20 Coordinated prefetching based on training in hierarchically cached processors

Publications (2)

Publication Number Publication Date
EP2642398A1 EP2642398A1 (fr) 2013-09-25
EP2642398B1 true EP2642398B1 (fr) 2015-05-20

Family

ID=47900909

Family Applications (1)

Application Number Title Priority Date Filing Date
EP20130159754 Active EP2642398B1 (fr) 2012-03-20 2013-03-18 Prélecture coordonnée dans des processeurs en mémoire cache hiérarchique

Country Status (6)

Country Link
US (1) US9098418B2 (fr)
EP (1) EP2642398B1 (fr)
KR (1) KR101483849B1 (fr)
CN (1) CN103324585B (fr)
TW (1) TWI545435B (fr)
WO (1) WO2013142148A2 (fr)

Families Citing this family (51)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150334016A1 (en) * 2009-11-26 2015-11-19 Nec Corporation Relay device
US8966185B2 (en) 2012-06-14 2015-02-24 International Business Machines Corporation Cache memory prefetching
US9348754B2 (en) 2012-10-11 2016-05-24 Soft Machines Inc. Systems and methods for implementing weak stream software data and instruction prefetching using a hardware data prefetcher
US9424046B2 (en) 2012-10-11 2016-08-23 Soft Machines Inc. Systems and methods for load canceling in a processor that is connected to an external interconnect fabric
US9483406B2 (en) * 2013-03-11 2016-11-01 Via Technologies, Inc. Communicating prefetchers that throttle one another
US20150095586A1 (en) * 2013-09-30 2015-04-02 Advanced Micro Devices , Inc. Storing non-temporal cache data
US9569361B2 (en) 2014-01-10 2017-02-14 Samsung Electronics Co., Ltd. Pre-fetch chaining
CN103747105A (zh) * 2014-01-26 2014-04-23 深圳市远行科技有限公司 一种网络文件的缓存方法及系统
US9811467B2 (en) * 2014-02-03 2017-11-07 Cavium, Inc. Method and an apparatus for pre-fetching and processing work for procesor cores in a network processor
US9436786B1 (en) * 2014-02-12 2016-09-06 Xilinx, Inc. Method and circuits for superclocking
CN104809080B (zh) * 2014-05-27 2019-09-17 威盛电子股份有限公司 相互节流的通信预取器
US9612970B2 (en) * 2014-07-17 2017-04-04 Qualcomm Incorporated Method and apparatus for flexible cache partitioning by sets and ways into component caches
US20160041914A1 (en) * 2014-08-05 2016-02-11 Advanced Micro Devices, Inc. Cache Bypassing Policy Based on Prefetch Streams
EP3198838A1 (fr) * 2014-09-25 2017-08-02 Interdigital Patent Holdings, Inc. Procédures de mise en cache sensible au contenu et de gestion de ressources radio pour transmission coordonnée multipoint
US9891916B2 (en) * 2014-10-20 2018-02-13 Via Technologies, Inc. Dynamically updating hardware prefetch trait to exclusive or shared in multi-memory access agent system
CN104378420B (zh) * 2014-11-12 2017-12-29 东软集团股份有限公司 基于环境感知的数据传输方法及装置
US10387318B2 (en) * 2014-12-14 2019-08-20 Via Alliance Semiconductor Co., Ltd Prefetching with level of aggressiveness based on effectiveness by memory access type
WO2016097809A1 (fr) * 2014-12-14 2016-06-23 Via Alliance Semiconductor Co., Ltd. Pré-extracteurs de données multiples qui défèrent l'un vers l'autre en fonction de l'efficacité de pré-extraction par type d'accès de mémoire
US9734072B2 (en) * 2015-03-24 2017-08-15 Macom Connectivity Solutions, Llc Main memory prefetch operation and multiple prefetch operation
US9600417B2 (en) * 2015-04-29 2017-03-21 Google Inc. Data caching
US9971694B1 (en) 2015-06-24 2018-05-15 Apple Inc. Prefetch circuit for a processor with pointer optimization
US11169925B2 (en) * 2015-08-25 2021-11-09 Samsung Electronics Co., Ltd. Capturing temporal store streams into CPU caches by dynamically varying store streaming thresholds
US10642617B2 (en) 2015-12-08 2020-05-05 Via Alliance Semiconductor Co., Ltd. Processor with an expandable instruction set architecture for dynamically configuring execution resources
US11061853B2 (en) 2015-12-08 2021-07-13 Via Alliance Semiconductor Co., Ltd. Processor with memory controller including dynamically programmable functional unit
US10268586B2 (en) * 2015-12-08 2019-04-23 Via Alliance Semiconductor Co., Ltd. Processor with programmable prefetcher operable to generate at least one prefetch address based on load requests
US9934149B2 (en) 2016-03-31 2018-04-03 Qualcomm Incorporated Prefetch mechanism for servicing demand miss
US10180905B1 (en) 2016-04-07 2019-01-15 Apple Inc. Unified prefetch circuit for multi-level caches
US9904624B1 (en) 2016-04-07 2018-02-27 Apple Inc. Prefetch throttling in a multi-core system
US10013357B2 (en) 2016-05-09 2018-07-03 Cavium, Inc. Managing memory access requests with prefetch for streams
US10452551B2 (en) * 2016-12-12 2019-10-22 Intel Corporation Programmable memory prefetcher for prefetching multiple cache lines based on data in a prefetch engine control register
US10242654B2 (en) * 2017-01-25 2019-03-26 Microsoft Technology Licensing, Llc No miss cache structure for real-time image transformations
US9978118B1 (en) * 2017-01-25 2018-05-22 Microsoft Technology Licensing, Llc No miss cache structure for real-time image transformations with data compression
US10331567B1 (en) 2017-02-17 2019-06-25 Apple Inc. Prefetch circuit with global quality factor to reduce aggressiveness in low power modes
US10514753B2 (en) 2017-03-27 2019-12-24 Microsoft Technology Licensing, Llc Selectively applying reprojection processing to multi-layer scenes for optimizing late stage reprojection power
US10410349B2 (en) 2017-03-27 2019-09-10 Microsoft Technology Licensing, Llc Selective application of reprojection processing on layer sub-regions for optimizing late stage reprojection power
US10255891B2 (en) 2017-04-12 2019-04-09 Microsoft Technology Licensing, Llc No miss cache structure for real-time image transformations with multiple LSR processing engines
US10387320B2 (en) 2017-05-12 2019-08-20 Samsung Electronics Co., Ltd. Integrated confirmation queues
US10613983B2 (en) * 2018-03-20 2020-04-07 Advanced Micro Devices, Inc. Prefetcher based speculative dynamic random-access memory read request technique
CN109446111B (zh) * 2018-10-15 2021-01-26 上海兆芯集成电路有限公司 存储器集成电路及其预取地址决定方法
CN110232030B (zh) * 2019-06-12 2021-08-10 上海兆芯集成电路有限公司 多芯片系统及缓存处理方法
US11126556B1 (en) 2019-07-19 2021-09-21 Marvell Asia Pte, Ltd. History table management for a correlated prefetcher
CN110865947B (zh) * 2019-11-14 2022-02-08 中国人民解放军国防科技大学 一种面向预取数据的高速缓存管理方法
CN113128531B (zh) * 2019-12-30 2024-03-26 上海商汤智能科技有限公司 一种数据处理方法及装置
CN113435601A (zh) * 2020-03-23 2021-09-24 华为技术有限公司 数据预取方法、装置以及存储设备
US11442864B2 (en) 2020-06-29 2022-09-13 Marvell Asia Pte, Ltd. Managing prefetch requests based on stream information for previously recognized streams
CN113568848B (zh) * 2020-07-29 2023-07-11 华为技术有限公司 处理器、信号调整方法及计算机系统
KR20220127504A (ko) * 2021-03-11 2022-09-20 삼성전자주식회사 블록 장치 이미지에 대한 런타임 체크섬 검증을 수행하는 전자 장치 및 이의 동작 방법
US20230306031A1 (en) * 2021-06-23 2023-09-28 Beijing Baidu Netcom Science Technology Co., Ltd. Method for data processing, computing device, and storage medium
US11630670B2 (en) 2021-07-21 2023-04-18 Apple Inc. Multi-table signature prefetch
CN114358179A (zh) * 2021-12-31 2022-04-15 海光信息技术股份有限公司 处理器的预取训练方法、处理装置、处理器和计算设备
US11940921B2 (en) * 2022-01-07 2024-03-26 Centaur Technology, Inc. Bounding box prefetcher

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6381678B2 (en) * 1998-10-30 2002-04-30 Intel Corporation Processing ordered data requests to a memory
US6519682B2 (en) 1998-12-04 2003-02-11 Stmicroelectronics, Inc. Pipelined non-blocking level two cache system with inherent transaction collision-avoidance
US6317811B1 (en) 1999-08-26 2001-11-13 International Business Machines Corporation Method and system for reissuing load requests in a multi-stream prefetch design
US6446167B1 (en) * 1999-11-08 2002-09-03 International Business Machines Corporation Cache prefetching of L2 and L3
US6578130B2 (en) 2001-10-18 2003-06-10 International Business Machines Corporation Programmable data prefetch pacing
TWI227402B (en) * 2002-06-18 2005-02-01 Ip First Llc Microprocessor and method for performing selective prefetch based on bus activity level
US7836259B1 (en) 2004-04-02 2010-11-16 Advanced Micro Devices, Inc. Prefetch unit for use with a cache memory subsystem of a cache memory hierarchy
US8490065B2 (en) * 2005-10-13 2013-07-16 International Business Machines Corporation Method and apparatus for software-assisted data cache and prefetch control
US7774578B2 (en) 2006-06-07 2010-08-10 Advanced Micro Devices, Inc. Apparatus and method of prefetching data in response to a cache miss
US7937532B2 (en) 2007-03-30 2011-05-03 Intel Corporation Method and apparatus for speculative prefetching in a multi-processor/multi-core message-passing machine
US8255631B2 (en) * 2008-02-01 2012-08-28 International Business Machines Corporation Priority-based prefetch requests scheduling and throttling
US8266393B2 (en) 2008-06-04 2012-09-11 Microsoft Corporation Coordination among multiple memory controllers
US8140769B2 (en) 2009-04-20 2012-03-20 Oracle America, Inc. Data prefetcher
US8667225B2 (en) * 2009-09-11 2014-03-04 Advanced Micro Devices, Inc. Store aware prefetching for a datastream
US8769209B2 (en) * 2010-12-20 2014-07-01 Intel Corporation Method and apparatus for achieving non-inclusive cache performance with inclusive caches
US9026739B2 (en) * 2012-03-07 2015-05-05 Advanced Micro Devices, Inc. Multimode prefetcher

Also Published As

Publication number Publication date
TWI545435B (zh) 2016-08-11
KR20130106789A (ko) 2013-09-30
WO2013142148A3 (fr) 2015-07-02
CN103324585B (zh) 2016-08-10
CN103324585A (zh) 2013-09-25
WO2013142148A2 (fr) 2013-09-26
US9098418B2 (en) 2015-08-04
US20130254485A1 (en) 2013-09-26
EP2642398A1 (fr) 2013-09-25
TW201346556A (zh) 2013-11-16
KR101483849B1 (ko) 2015-01-16

Similar Documents

Publication Publication Date Title
EP2642398B1 (fr) Prélecture coordonnée dans des processeurs en mémoire cache hiérarchique
US9047198B2 (en) Prefetching across page boundaries in hierarchically cached processors
US10394718B2 (en) Slot/sub-slot prefetch architecture for multiple memory requestors
US8621157B2 (en) Cache prefetching from non-uniform memories
US8751746B2 (en) QoS management in the L2 cache
EP2786255B1 (fr) Mémoire cache dram avec balises et données conjointement stockées dans des rangées physiques
US8868843B2 (en) Hardware filter for tracking block presence in large caches
US20120221785A1 (en) Polymorphic Stacked DRAM Memory Architecture
US8996815B2 (en) Cache memory controller
US9552301B2 (en) Method and apparatus related to cache memory
KR20120070584A (ko) 데이터 스트림에 대한 저장 인식 프리페치
US9043554B2 (en) Cache policies for uncacheable memory requests
US9043570B2 (en) System cache with quota-based control
WO2014052383A1 (fr) Mémoire cache de système avec état d'attente de données
US9229866B2 (en) Delaying cache data array updates
EP3844624B1 (fr) Procédé, appareil et système de réduction d'arrêts de pipeline dus à des absences de traductions d'adresses
EP2901287B1 (fr) Cache système, avec moteur de suppression de verrouillage
US20090006777A1 (en) Apparatus for reducing cache latency while preserving cache bandwidth in a cache subsystem of a processor
US10963392B1 (en) Victim allocations in shared system cache
US10725928B1 (en) Translation lookaside buffer invalidation by range
US20120131305A1 (en) Page aware prefetch mechanism
US11645207B2 (en) Prefetch disable of memory requests targeting data lacking locality

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20130318

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

17Q First examination report despatched

Effective date: 20140221

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

INTG Intention to grant announced

Effective date: 20141205

GRAS Grant fee paid

Free format text: ORIGINAL CODE: EPIDOSNIGR3

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: CH

Ref legal event code: EP

REG Reference to a national code

Ref country code: AT

Ref legal event code: REF

Ref document number: 728048

Country of ref document: AT

Kind code of ref document: T

Effective date: 20150615

REG Reference to a national code

Ref country code: IE

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: DE

Ref legal event code: R096

Ref document number: 602013001779

Country of ref document: DE

Effective date: 20150702

REG Reference to a national code

Ref country code: AT

Ref legal event code: MK05

Ref document number: 728048

Country of ref document: AT

Kind code of ref document: T

Effective date: 20150520

REG Reference to a national code

Ref country code: LT

Ref legal event code: MG4D

REG Reference to a national code

Ref country code: NL

Ref legal event code: MP

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: HR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: NO

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150820

Ref country code: PT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150921

Ref country code: ES

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: FI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: LT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: IS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150920

Ref country code: GR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150821

Ref country code: AT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: BG

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150820

Ref country code: RS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: LV

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: DK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: EE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

REG Reference to a national code

Ref country code: DE

Ref legal event code: R097

Ref document number: 602013001779

Country of ref document: DE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: CZ

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: RO

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20150520

Ref country code: PL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: SK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

26N No opposition filed

Effective date: 20160223

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: BE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: MC

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: LU

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20160318

REG Reference to a national code

Ref country code: CH

Ref legal event code: PL

REG Reference to a national code

Ref country code: IE

Ref legal event code: MM4A

REG Reference to a national code

Ref country code: FR

Ref legal event code: ST

Effective date: 20161130

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: LI

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20160331

Ref country code: IE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20160318

Ref country code: FR

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20160331

Ref country code: CH

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20160331

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: SE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: MT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: HU

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT; INVALID AB INITIO

Effective date: 20130318

Ref country code: CY

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: SM

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: MK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

Ref country code: MT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20160331

Ref country code: TR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: AL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20150520

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 20230126

Year of fee payment: 11

Ref country code: DE

Payment date: 20230125

Year of fee payment: 11

P01 Opt-out of the competence of the unified patent court (upc) registered

Effective date: 20230523